Politics aside, I've always enjoyed watching the occasional French president's speech on television. Where U.S. presidents act, well, presidential, and the Queen will be regal, a French head of state is much more expansive. And nowadays, we can also compare another style: on-line presence.

Obama, of course, is on YouTube; and YouTube started an experiment with automatic closed captioning of videos last year, so some of Obama's speeches have automatically transcribed subtitles.

Sarkozy, on the other hand, resides online on the much more presidential Elysée site. Even if this weren't a perfect case in point for "not invented here," YouTube doesn't understand anything but English, so that wasn't an option. Instead, the French search vendor Exalead provided text-to-speech for sous titrage, and made all the videos keyword searchable. The results point to the exact frame in the video, too.

Of course, the idea itself isn't new. A few vendors in our Search & Information Access research allow you to do the same. If you'd want to search loads of untranscribed recordings -- say, council meetings, or millions of customer recordings from a call center -- you could ask Autonomy or Coveo as well. (Autonomy's technology also powers the Blinkx video search, and Coveo has a module you can also use to enable SharePoint to index audio and video.)

However, if there's ever a technology you'd have to test on your own corpus first, before purchasing, media search is it. The success rate varies wildly, and the most important factor isn't even the technology -- it's the quality of the recording. Exalead's Voxalead has been trained to understand French political speeches, and the Sarkozy recordings are impeccable, so there are relatively few mistakes. The Voxalead News demo is a lot more ambitious (indexing newscasts in five languages), but doesn't fare quite as well, even though the sound quality is usually pretty good. Do this over the phone and speak in vernacular, and results can range from surprisingly good to useless -- but funny. (Note that while YouTube and Google Voice both use the same technology, it's not available as a module for the Google Appliance.)

So though I can think of quite a few uses for the technology, in most cases, don't hold your breath yet. The demos can be amazing, and Exalead proves it's a lot better at keeping tabs on Sarkozy than Google is with Obama. But before you commit, test, test again, and if you think you're done, do more testing. Truly useful and successful implementations are still few and far between.